1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_riscv.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/resolvedFieldEntry.hpp"
40 #include "oops/resolvedIndyEntry.hpp"
41 #include "oops/resolvedMethodEntry.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "prims/jvmtiThreadState.hpp"
44 #include "runtime/basicLock.hpp"
45 #include "runtime/frame.inline.hpp"
46 #include "runtime/javaThread.hpp"
47 #include "runtime/safepointMechanism.hpp"
48 #include "runtime/sharedRuntime.hpp"
49 #include "utilities/powerOfTwo.hpp"
50
51 void InterpreterMacroAssembler::narrow(Register result) {
52 // Get method->_constMethod->_result_type
53 ld(t0, Address(fp, frame::interpreter_frame_method_offset * wordSize));
54 ld(t0, Address(t0, Method::const_offset()));
55 lbu(t0, Address(t0, ConstMethod::result_type_offset()));
56
57 Label done, notBool, notByte, notChar;
58
59 // common case first
60 mv(t1, T_INT);
61 beq(t0, t1, done);
62
63 // mask integer result to narrower return type.
64 mv(t1, T_BOOLEAN);
65 bne(t0, t1, notBool);
66
67 andi(result, result, 0x1);
68 j(done);
69
70 bind(notBool);
71 mv(t1, T_BYTE);
72 bne(t0, t1, notByte);
73 sign_extend(result, result, 8);
74 j(done);
75
76 bind(notByte);
77 mv(t1, T_CHAR);
78 bne(t0, t1, notChar);
79 zero_extend(result, result, 16);
80 j(done);
81
82 bind(notChar);
83 sign_extend(result, result, 16);
84
85 bind(done);
86 sign_extend(result, result, 32);
87 }
88
89 void InterpreterMacroAssembler::jump_to_entry(address entry) {
90 assert(entry != nullptr, "Entry must have been generated by now");
91 j(entry);
92 }
93
94 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
95 if (JvmtiExport::can_pop_frame()) {
96 Label L;
97 // Initiate popframe handling only if it is not already being
98 // processed. If the flag has the popframe_processing bit set,
99 // it means that this code is called *during* popframe handling - we
100 // don't want to reenter.
101 // This method is only called just after the call into the vm in
102 // call_VM_base, so the arg registers are available.
103 lwu(t1, Address(xthread, JavaThread::popframe_condition_offset()));
104 test_bit(t0, t1, exact_log2(JavaThread::popframe_pending_bit));
105 beqz(t0, L);
106 test_bit(t0, t1, exact_log2(JavaThread::popframe_processing_bit));
107 bnez(t0, L);
108 // Call Interpreter::remove_activation_preserving_args_entry() to get the
109 // address of the same-named entrypoint in the generated interpreter code.
110 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
111 jr(x10);
112 bind(L);
113 }
114 }
115
116
117 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
118 ld(x12, Address(xthread, JavaThread::jvmti_thread_state_offset()));
119 const Address tos_addr(x12, JvmtiThreadState::earlyret_tos_offset());
120 const Address oop_addr(x12, JvmtiThreadState::earlyret_oop_offset());
121 const Address val_addr(x12, JvmtiThreadState::earlyret_value_offset());
122 switch (state) {
123 case atos:
124 ld(x10, oop_addr);
125 sd(zr, oop_addr);
126 verify_oop(x10);
127 break;
128 case ltos:
129 ld(x10, val_addr);
130 break;
131 case btos: // fall through
132 case ztos: // fall through
133 case ctos: // fall through
134 case stos: // fall through
135 case itos:
136 lwu(x10, val_addr);
137 break;
138 case ftos:
139 flw(f10, val_addr);
140 break;
141 case dtos:
142 fld(f10, val_addr);
143 break;
144 case vtos:
145 /* nothing to do */
146 break;
147 default:
148 ShouldNotReachHere();
149 }
150 // Clean up tos value in the thread object
151 mv(t0, (int)ilgl);
152 sw(t0, tos_addr);
153 sw(zr, val_addr);
154 }
155
156
157 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
158 if (JvmtiExport::can_force_early_return()) {
159 Label L;
160 ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
161 beqz(t0, L); // if thread->jvmti_thread_state() is null then exit
162
163 // Initiate earlyret handling only if it is not already being processed.
164 // If the flag has the earlyret_processing bit set, it means that this code
165 // is called *during* earlyret handling - we don't want to reenter.
166 lwu(t0, Address(t0, JvmtiThreadState::earlyret_state_offset()));
167 mv(t1, JvmtiThreadState::earlyret_pending);
168 bne(t0, t1, L);
169
170 // Call Interpreter::remove_activation_early_entry() to get the address of the
171 // same-named entrypoint in the generated interpreter code.
172 ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
173 lwu(t0, Address(t0, JvmtiThreadState::earlyret_tos_offset()));
174 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), t0);
175 jr(x10);
176 bind(L);
177 }
178 }
179
180 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
181 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
182 if (AvoidUnalignedAccesses && (bcp_offset % 2)) {
183 lbu(t1, Address(xbcp, bcp_offset));
184 lbu(reg, Address(xbcp, bcp_offset + 1));
185 slli(t1, t1, 8);
186 add(reg, reg, t1);
187 } else {
188 lhu(reg, Address(xbcp, bcp_offset));
189 revb_h_h_u(reg, reg);
190 }
191 }
192
193 void InterpreterMacroAssembler::get_dispatch() {
194 ExternalAddress target((address)Interpreter::dispatch_table());
195 relocate(target.rspec(), [&] {
196 int32_t offset;
197 la(xdispatch, target.target(), offset);
198 addi(xdispatch, xdispatch, offset);
199 });
200 }
201
202 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
203 Register tmp,
204 int bcp_offset,
205 size_t index_size) {
206 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
207 if (index_size == sizeof(u2)) {
208 if (AvoidUnalignedAccesses) {
209 assert_different_registers(index, tmp);
210 load_unsigned_byte(index, Address(xbcp, bcp_offset));
211 load_unsigned_byte(tmp, Address(xbcp, bcp_offset + 1));
212 slli(tmp, tmp, 8);
213 add(index, index, tmp);
214 } else {
215 load_unsigned_short(index, Address(xbcp, bcp_offset));
216 }
217 } else if (index_size == sizeof(u4)) {
218 load_int_misaligned(index, Address(xbcp, bcp_offset), tmp, false);
219 } else if (index_size == sizeof(u1)) {
220 load_unsigned_byte(index, Address(xbcp, bcp_offset));
221 } else {
222 ShouldNotReachHere();
223 }
224 }
225
226 // Load object from cpool->resolved_references(index)
227 void InterpreterMacroAssembler::load_resolved_reference_at_index(
228 Register result, Register index, Register tmp) {
229 assert_different_registers(result, index);
230
231 get_constant_pool(result);
232 // Load pointer for resolved_references[] objArray
233 ld(result, Address(result, ConstantPool::cache_offset()));
234 ld(result, Address(result, ConstantPoolCache::resolved_references_offset()));
235 resolve_oop_handle(result, tmp, t1);
236 // Add in the index
237 addi(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
238 shadd(result, index, result, index, LogBytesPerHeapOop);
239 load_heap_oop(result, Address(result, 0), tmp, t1);
240 }
241
242 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
243 Register cpool, Register index, Register klass, Register temp) {
244 shadd(temp, index, cpool, temp, LogBytesPerWord);
245 lhu(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
246 ld(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
247 shadd(klass, temp, klass, temp, LogBytesPerWord);
248 ld(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
249 }
250
251 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
252 // subtype of super_klass.
253 //
254 // Args:
255 // x10: superklass
256 // Rsub_klass: subklass
257 //
258 // Kills:
259 // x12, x15
260 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
261 Label& ok_is_subtype) {
262 assert(Rsub_klass != x10, "x10 holds superklass");
263 assert(Rsub_klass != x12, "x12 holds 2ndary super array length");
264 assert(Rsub_klass != x15, "x15 holds 2ndary super array scan ptr");
265
266 // Profile the not-null value's klass.
267 profile_typecheck(x12, Rsub_klass, x15); // blows x12, reloads x15
268
269 // Do the check.
270 check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12
271 }
272
273 // Java Expression Stack
274
275 void InterpreterMacroAssembler::pop_ptr(Register r) {
276 ld(r, Address(esp, 0));
277 addi(esp, esp, wordSize);
278 }
279
280 void InterpreterMacroAssembler::pop_i(Register r) {
281 lw(r, Address(esp, 0)); // lw do signed extended
282 addi(esp, esp, wordSize);
283 }
284
285 void InterpreterMacroAssembler::pop_l(Register r) {
286 ld(r, Address(esp, 0));
287 addi(esp, esp, 2 * Interpreter::stackElementSize);
288 }
289
290 void InterpreterMacroAssembler::push_ptr(Register r) {
291 addi(esp, esp, -wordSize);
292 sd(r, Address(esp, 0));
293 }
294
295 void InterpreterMacroAssembler::push_i(Register r) {
296 addi(esp, esp, -wordSize);
297 sign_extend(r, r, 32);
298 sd(r, Address(esp, 0));
299 }
300
301 void InterpreterMacroAssembler::push_l(Register r) {
302 addi(esp, esp, -2 * wordSize);
303 sd(zr, Address(esp, wordSize));
304 sd(r, Address(esp));
305 }
306
307 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
308 flw(r, Address(esp, 0));
309 addi(esp, esp, wordSize);
310 }
311
312 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
313 fld(r, Address(esp, 0));
314 addi(esp, esp, 2 * Interpreter::stackElementSize);
315 }
316
317 void InterpreterMacroAssembler::push_f(FloatRegister r) {
318 addi(esp, esp, -wordSize);
319 fsw(r, Address(esp, 0));
320 }
321
322 void InterpreterMacroAssembler::push_d(FloatRegister r) {
323 addi(esp, esp, -2 * wordSize);
324 fsd(r, Address(esp, 0));
325 }
326
327 void InterpreterMacroAssembler::pop(TosState state) {
328 switch (state) {
329 case atos:
330 pop_ptr();
331 verify_oop(x10);
332 break;
333 case btos: // fall through
334 case ztos: // fall through
335 case ctos: // fall through
336 case stos: // fall through
337 case itos:
338 pop_i();
339 break;
340 case ltos:
341 pop_l();
342 break;
343 case ftos:
344 pop_f();
345 break;
346 case dtos:
347 pop_d();
348 break;
349 case vtos:
350 /* nothing to do */
351 break;
352 default:
353 ShouldNotReachHere();
354 }
355 }
356
357 void InterpreterMacroAssembler::push(TosState state) {
358 switch (state) {
359 case atos:
360 verify_oop(x10);
361 push_ptr();
362 break;
363 case btos: // fall through
364 case ztos: // fall through
365 case ctos: // fall through
366 case stos: // fall through
367 case itos:
368 push_i();
369 break;
370 case ltos:
371 push_l();
372 break;
373 case ftos:
374 push_f();
375 break;
376 case dtos:
377 push_d();
378 break;
379 case vtos:
380 /* nothing to do */
381 break;
382 default:
383 ShouldNotReachHere();
384 }
385 }
386
387 // Helpers for swap and dup
388 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
389 ld(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
390 }
391
392 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
393 sd(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
394 }
395
396 void InterpreterMacroAssembler::load_float(Address src) {
397 flw(f10, src);
398 }
399
400 void InterpreterMacroAssembler::load_double(Address src) {
401 fld(f10, src);
402 }
403
404 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
405 // set sender sp
406 mv(x19_sender_sp, sp);
407 // record last_sp
408 sub(t0, esp, fp);
409 srai(t0, t0, Interpreter::logStackElementSize);
410 sd(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
411 }
412
413 // Jump to from_interpreted entry of a call unless single stepping is possible
414 // in this thread in which case we must call the i2i entry
415 void InterpreterMacroAssembler::jump_from_interpreted(Register method) {
416 prepare_to_jump_from_interpreted();
417 if (JvmtiExport::can_post_interpreter_events()) {
418 Label run_compiled_code;
419 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
420 // compiled code in threads for which the event is enabled. Check here for
421 // interp_only_mode if these events CAN be enabled.
422 lwu(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
423 beqz(t0, run_compiled_code);
424 ld(t0, Address(method, Method::interpreter_entry_offset()));
425 jr(t0);
426 bind(run_compiled_code);
427 }
428
429 ld(t0, Address(method, Method::from_interpreted_offset()));
430 jr(t0);
431 }
432
433 // The following two routines provide a hook so that an implementation
434 // can schedule the dispatch in two parts. amd64 does not do this.
435 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
436 }
437
438 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
439 dispatch_next(state, step);
440 }
441
442 void InterpreterMacroAssembler::dispatch_base(TosState state,
443 address* table,
444 bool verifyoop,
445 bool generate_poll,
446 Register Rs) {
447 // Pay attention to the argument Rs, which is acquiesce in t0.
448 if (VerifyActivationFrameSize) {
449 Unimplemented();
450 }
451 if (verifyoop && state == atos) {
452 verify_oop(x10);
453 }
454
455 Label safepoint;
456 address* const safepoint_table = Interpreter::safept_table(state);
457 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
458
459 if (needs_thread_local_poll) {
460 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
461 ld(t1, Address(xthread, JavaThread::polling_word_offset()));
462 test_bit(t1, t1, exact_log2(SafepointMechanism::poll_bit()));
463 bnez(t1, safepoint);
464 }
465 if (table == Interpreter::dispatch_table(state)) {
466 mv(t1, Interpreter::distance_from_dispatch_table(state));
467 add(t1, Rs, t1);
468 shadd(t1, t1, xdispatch, t1, 3);
469 } else {
470 mv(t1, (address)table);
471 shadd(t1, Rs, t1, Rs, 3);
472 }
473 ld(t1, Address(t1));
474 jr(t1);
475
476 if (needs_thread_local_poll) {
477 bind(safepoint);
478 la(t1, ExternalAddress((address)safepoint_table));
479 shadd(t1, Rs, t1, Rs, 3);
480 ld(t1, Address(t1));
481 jr(t1);
482 }
483 }
484
485 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll, Register Rs) {
486 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll, Rs);
487 }
488
489 void InterpreterMacroAssembler::dispatch_only_normal(TosState state, Register Rs) {
490 dispatch_base(state, Interpreter::normal_table(state), true, false, Rs);
491 }
492
493 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state, Register Rs) {
494 dispatch_base(state, Interpreter::normal_table(state), false, false, Rs);
495 }
496
497 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
498 // load next bytecode
499 load_unsigned_byte(t0, Address(xbcp, step));
500 add(xbcp, xbcp, step);
501 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
502 }
503
504 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
505 // load current bytecode
506 lbu(t0, Address(xbcp, 0));
507 dispatch_base(state, table);
508 }
509
510 // remove activation
511 //
512 // Apply stack watermark barrier.
513 // Unlock the receiver if this is a synchronized method.
514 // Unlock any Java monitors from synchronized blocks.
515 // Remove the activation from the stack.
516 //
517 // If there are locked Java monitors
518 // If throw_monitor_exception
519 // throws IllegalMonitorStateException
520 // Else if install_monitor_exception
521 // installs IllegalMonitorStateException
522 // Else
523 // no error processing
524 void InterpreterMacroAssembler::remove_activation(
525 TosState state,
526 bool throw_monitor_exception,
527 bool install_monitor_exception,
528 bool notify_jvmdi) {
529 // Note: Registers x13 may be in use for the
530 // result check if synchronized method
531 Label unlocked, unlock, no_unlock;
532
533 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
534 // that would normally not be safe to use. Such bad returns into unsafe territory of
535 // the stack, will call InterpreterRuntime::at_unwind.
536 Label slow_path;
537 Label fast_path;
538 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
539 j(fast_path);
540
541 bind(slow_path);
542 push(state);
543 set_last_Java_frame(esp, fp, (address)pc(), t0);
544 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), xthread);
545 reset_last_Java_frame(true);
546 pop(state);
547
548 bind(fast_path);
549
550 // get the value of _do_not_unlock_if_synchronized into x13
551 const Address do_not_unlock_if_synchronized(xthread,
552 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
553 lbu(x13, do_not_unlock_if_synchronized);
554 sb(zr, do_not_unlock_if_synchronized); // reset the flag
555
556 // get method access flags
557 ld(x11, Address(fp, frame::interpreter_frame_method_offset * wordSize));
558 ld(x12, Address(x11, Method::access_flags_offset()));
559 test_bit(t0, x12, exact_log2(JVM_ACC_SYNCHRONIZED));
560 beqz(t0, unlocked);
561
562 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
563 // is set.
564 bnez(x13, no_unlock);
565
566 // unlock monitor
567 push(state); // save result
568
569 // BasicObjectLock will be first in list, since this is a
570 // synchronized method. However, need to check that the object has
571 // not been unlocked by an explicit monitorexit bytecode.
572 const Address monitor(fp, frame::interpreter_frame_initial_sp_offset *
573 wordSize - (int) sizeof(BasicObjectLock));
574 // We use c_rarg1 so that if we go slow path it will be the correct
575 // register for unlock_object to pass to VM directly
576 la(c_rarg1, monitor); // address of first monitor
577
578 ld(x10, Address(c_rarg1, BasicObjectLock::obj_offset()));
579 bnez(x10, unlock);
580
581 pop(state);
582 if (throw_monitor_exception) {
583 // Entry already unlocked, need to throw exception
584 call_VM(noreg, CAST_FROM_FN_PTR(address,
585 InterpreterRuntime::throw_illegal_monitor_state_exception));
586 should_not_reach_here();
587 } else {
588 // Monitor already unlocked during a stack unroll. If requested,
589 // install an illegal_monitor_state_exception. Continue with
590 // stack unrolling.
591 if (install_monitor_exception) {
592 call_VM(noreg, CAST_FROM_FN_PTR(address,
593 InterpreterRuntime::new_illegal_monitor_state_exception));
594 }
595 j(unlocked);
596 }
597
598 bind(unlock);
599 unlock_object(c_rarg1);
600 pop(state);
601
602 // Check that for block-structured locking (i.e., that all locked
603 // objects has been unlocked)
604 bind(unlocked);
605
606 // x10: Might contain return value
607
608 // Check that all monitors are unlocked
609 {
610 Label loop, exception, entry, restart;
611 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
612 const Address monitor_block_top(
613 fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
614 const Address monitor_block_bot(
615 fp, frame::interpreter_frame_initial_sp_offset * wordSize);
616
617 bind(restart);
618 // We use c_rarg1 so that if we go slow path it will be the correct
619 // register for unlock_object to pass to VM directly
620 ld(c_rarg1, monitor_block_top); // derelativize pointer
621 shadd(c_rarg1, c_rarg1, fp, c_rarg1, LogBytesPerWord);
622 // c_rarg1 points to current entry, starting with top-most entry
623
624 la(x9, monitor_block_bot); // points to word before bottom of
625 // monitor block
626
627 j(entry);
628
629 // Entry already locked, need to throw exception
630 bind(exception);
631
632 if (throw_monitor_exception) {
633 // Throw exception
634 MacroAssembler::call_VM(noreg,
635 CAST_FROM_FN_PTR(address, InterpreterRuntime::
636 throw_illegal_monitor_state_exception));
637
638 should_not_reach_here();
639 } else {
640 // Stack unrolling. Unlock object and install illegal_monitor_exception.
641 // Unlock does not block, so don't have to worry about the frame.
642 // We don't have to preserve c_rarg1 since we are going to throw an exception.
643
644 push(state);
645 unlock_object(c_rarg1);
646 pop(state);
647
648 if (install_monitor_exception) {
649 call_VM(noreg, CAST_FROM_FN_PTR(address,
650 InterpreterRuntime::
651 new_illegal_monitor_state_exception));
652 }
653
654 j(restart);
655 }
656
657 bind(loop);
658 // check if current entry is used
659 add(t0, c_rarg1, in_bytes(BasicObjectLock::obj_offset()));
660 ld(t0, Address(t0, 0));
661 bnez(t0, exception);
662
663 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
664 bind(entry);
665 bne(c_rarg1, x9, loop); // check if bottom reached if not at bottom then check this entry
666 }
667
668 bind(no_unlock);
669
670 // jvmti support
671 if (notify_jvmdi) {
672 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
673
674 } else {
675 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
676 }
677
678 // remove activation
679 // get sender esp
680 ld(t1,
681 Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize));
682 if (StackReservedPages > 0) {
683 // testing if reserved zone needs to be re-enabled
684 Label no_reserved_zone_enabling;
685
686 // check if already enabled - if so no re-enabling needed
687 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
688 lw(t0, Address(xthread, JavaThread::stack_guard_state_offset()));
689 subw(t0, t0, StackOverflow::stack_guard_enabled);
690 beqz(t0, no_reserved_zone_enabling);
691
692 ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset()));
693 ble(t1, t0, no_reserved_zone_enabling);
694
695 call_VM_leaf(
696 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), xthread);
697 call_VM(noreg, CAST_FROM_FN_PTR(address,
698 InterpreterRuntime::throw_delayed_StackOverflowError));
699 should_not_reach_here();
700
701 bind(no_reserved_zone_enabling);
702 }
703
704 // restore sender esp
705 mv(esp, t1);
706
707 // remove frame anchor
708 leave();
709 // If we're returning to interpreted code we will shortly be
710 // adjusting SP to allow some space for ESP. If we're returning to
711 // compiled code the saved sender SP was saved in sender_sp, so this
712 // restores it.
713 andi(sp, esp, -16);
714 }
715
716 // Lock object
717 //
718 // Args:
719 // c_rarg1: BasicObjectLock to be used for locking
720 //
721 // Kills:
722 // x10
723 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5, .. (param regs)
724 // t0, t1 (temp regs)
725 void InterpreterMacroAssembler::lock_object(Register lock_reg)
726 {
727 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
728 if (LockingMode == LM_MONITOR) {
729 call_VM_preemptable(noreg,
730 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
731 lock_reg);
732 } else {
733 Label count, done;
734
735 const Register swap_reg = x10;
736 const Register tmp = c_rarg2;
737 const Register obj_reg = c_rarg3; // Will contain the oop
738 const Register tmp2 = c_rarg4;
739 const Register tmp3 = c_rarg5;
740
741 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
742 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
743 const int mark_offset = lock_offset +
744 BasicLock::displaced_header_offset_in_bytes();
745
746 Label slow_case;
747
748 // Load object pointer into obj_reg c_rarg3
749 ld(obj_reg, Address(lock_reg, obj_offset));
750
751 if (DiagnoseSyncOnValueBasedClasses != 0) {
752 load_klass(tmp, obj_reg);
753 lbu(tmp, Address(tmp, Klass::misc_flags_offset()));
754 test_bit(tmp, tmp, exact_log2(KlassFlags::_misc_is_value_based_class));
755 bnez(tmp, slow_case);
756 }
757
758 if (LockingMode == LM_LIGHTWEIGHT) {
759 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
760 j(done);
761 } else if (LockingMode == LM_LEGACY) {
762 // Load (object->mark() | 1) into swap_reg
763 ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
764 ori(swap_reg, t0, 1);
765
766 // Save (object->mark() | 1) into BasicLock's displaced header
767 sd(swap_reg, Address(lock_reg, mark_offset));
768
769 assert(lock_offset == 0,
770 "displached header must be first word in BasicObjectLock");
771
772 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, tmp, count, /*fallthrough*/nullptr);
773
774 // Test if the oopMark is an obvious stack pointer, i.e.,
775 // 1) (mark & 7) == 0, and
776 // 2) sp <= mark < mark + os::pagesize()
777 //
778 // These 3 tests can be done by evaluating the following
779 // expression: ((mark - sp) & (7 - os::vm_page_size())),
780 // assuming both stack pointer and pagesize have their
781 // least significant 3 bits clear.
782 // NOTE: the oopMark is in swap_reg x10 as the result of cmpxchg
783 sub(swap_reg, swap_reg, sp);
784 mv(t0, (int64_t)(7 - (int)os::vm_page_size()));
785 andr(swap_reg, swap_reg, t0);
786
787 // Save the test result, for recursive case, the result is zero
788 sd(swap_reg, Address(lock_reg, mark_offset));
789 bnez(swap_reg, slow_case);
790
791 bind(count);
792 inc_held_monitor_count();
793 j(done);
794 }
795
796 bind(slow_case);
797
798 // Call the runtime routine for slow case
799 call_VM_preemptable(noreg,
800 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
801 lock_reg);
802
803 bind(done);
804 }
805 }
806
807
808 // Unlocks an object. Used in monitorexit bytecode and
809 // remove_activation. Throws an IllegalMonitorException if object is
810 // not locked by current thread.
811 //
812 // Args:
813 // c_rarg1: BasicObjectLock for lock
814 //
815 // Kills:
816 // x10
817 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, ... (param regs)
818 // t0, t1 (temp regs)
819 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
820 {
821 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
822
823 if (LockingMode == LM_MONITOR) {
824 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
825 } else {
826 Label count, done;
827
828 const Register swap_reg = x10;
829 const Register header_reg = c_rarg2; // Will contain the old oopMark
830 const Register obj_reg = c_rarg3; // Will contain the oop
831 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
832
833 save_bcp(); // Save in case of exception
834
835 if (LockingMode != LM_LIGHTWEIGHT) {
836 // Convert from BasicObjectLock structure to object and BasicLock
837 // structure Store the BasicLock address into x10
838 la(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
839 }
840
841 // Load oop into obj_reg(c_rarg3)
842 ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
843
844 // Free entry
845 sd(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
846
847 Label slow_case;
848 if (LockingMode == LM_LIGHTWEIGHT) {
849 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
850 j(done);
851 } else if (LockingMode == LM_LEGACY) {
852 // Load the old header from BasicLock structure
853 ld(header_reg, Address(swap_reg,
854 BasicLock::displaced_header_offset_in_bytes()));
855
856 // Test for recursion
857 beqz(header_reg, count);
858
859 // Atomic swap back the old header
860 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, tmp_reg, count, &slow_case);
861
862 bind(count);
863 dec_held_monitor_count();
864 j(done);
865 }
866
867 bind(slow_case);
868 // Call the runtime routine for slow case.
869 sd(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
870 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
871
872 bind(done);
873 restore_bcp();
874 }
875 }
876
877
878 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
879 Label& zero_continue) {
880 assert(ProfileInterpreter, "must be profiling interpreter");
881 ld(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
882 beqz(mdp, zero_continue);
883 }
884
885 // Set the method data pointer for the current bcp.
886 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
887 assert(ProfileInterpreter, "must be profiling interpreter");
888 Label set_mdp;
889 push_reg(RegSet::of(x10, x11), sp); // save x10, x11
890
891 // Test MDO to avoid the call if it is null.
892 ld(x10, Address(xmethod, in_bytes(Method::method_data_offset())));
893 beqz(x10, set_mdp);
894 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), xmethod, xbcp);
895 // x10: mdi
896 // mdo is guaranteed to be non-zero here, we checked for it before the call.
897 ld(x11, Address(xmethod, in_bytes(Method::method_data_offset())));
898 la(x11, Address(x11, in_bytes(MethodData::data_offset())));
899 add(x10, x11, x10);
900 sd(x10, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
901 bind(set_mdp);
902 pop_reg(RegSet::of(x10, x11), sp);
903 }
904
905 void InterpreterMacroAssembler::verify_method_data_pointer() {
906 assert(ProfileInterpreter, "must be profiling interpreter");
907 #ifdef ASSERT
908 Label verify_continue;
909 add(sp, sp, -4 * wordSize);
910 sd(x10, Address(sp, 0));
911 sd(x11, Address(sp, wordSize));
912 sd(x12, Address(sp, 2 * wordSize));
913 sd(x13, Address(sp, 3 * wordSize));
914 test_method_data_pointer(x13, verify_continue); // If mdp is zero, continue
915 get_method(x11);
916
917 // If the mdp is valid, it will point to a DataLayout header which is
918 // consistent with the bcp. The converse is highly probable also.
919 lh(x12, Address(x13, in_bytes(DataLayout::bci_offset())));
920 ld(t0, Address(x11, Method::const_offset()));
921 add(x12, x12, t0);
922 la(x12, Address(x12, ConstMethod::codes_offset()));
923 beq(x12, xbcp, verify_continue);
924 // x10: method
925 // xbcp: bcp // xbcp == 22
926 // x13: mdp
927 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
928 x11, xbcp, x13);
929 bind(verify_continue);
930 ld(x10, Address(sp, 0));
931 ld(x11, Address(sp, wordSize));
932 ld(x12, Address(sp, 2 * wordSize));
933 ld(x13, Address(sp, 3 * wordSize));
934 add(sp, sp, 4 * wordSize);
935 #endif // ASSERT
936 }
937
938
939 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
940 int constant,
941 Register value) {
942 assert(ProfileInterpreter, "must be profiling interpreter");
943 Address data(mdp_in, constant);
944 sd(value, data);
945 }
946
947
948 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
949 int constant,
950 bool decrement) {
951 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
952 }
953
954 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
955 Register reg,
956 int constant,
957 bool decrement) {
958 assert(ProfileInterpreter, "must be profiling interpreter");
959 // %%% this does 64bit counters at best it is wasting space
960 // at worst it is a rare bug when counters overflow
961
962 assert_different_registers(t1, t0, mdp_in, reg);
963
964 Address addr1(mdp_in, constant);
965 Address addr2(t1, 0);
966 Address &addr = addr1;
967 if (reg != noreg) {
968 la(t1, addr1);
969 add(t1, t1, reg);
970 addr = addr2;
971 }
972
973 if (decrement) {
974 ld(t0, addr);
975 addi(t0, t0, -DataLayout::counter_increment);
976 Label L;
977 bltz(t0, L); // skip store if counter underflow
978 sd(t0, addr);
979 bind(L);
980 } else {
981 assert(DataLayout::counter_increment == 1,
982 "flow-free idiom only works with 1");
983 ld(t0, addr);
984 addi(t0, t0, DataLayout::counter_increment);
985 Label L;
986 blez(t0, L); // skip store if counter overflow
987 sd(t0, addr);
988 bind(L);
989 }
990 }
991
992 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
993 int flag_byte_constant) {
994 assert(ProfileInterpreter, "must be profiling interpreter");
995 int flags_offset = in_bytes(DataLayout::flags_offset());
996 // Set the flag
997 lbu(t1, Address(mdp_in, flags_offset));
998 ori(t1, t1, flag_byte_constant);
999 sb(t1, Address(mdp_in, flags_offset));
1000 }
1001
1002
1003 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1004 int offset,
1005 Register value,
1006 Register test_value_out,
1007 Label& not_equal_continue) {
1008 assert(ProfileInterpreter, "must be profiling interpreter");
1009 if (test_value_out == noreg) {
1010 ld(t1, Address(mdp_in, offset));
1011 bne(value, t1, not_equal_continue);
1012 } else {
1013 // Put the test value into a register, so caller can use it:
1014 ld(test_value_out, Address(mdp_in, offset));
1015 bne(value, test_value_out, not_equal_continue);
1016 }
1017 }
1018
1019
1020 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1021 int offset_of_disp) {
1022 assert(ProfileInterpreter, "must be profiling interpreter");
1023 ld(t1, Address(mdp_in, offset_of_disp));
1024 add(mdp_in, mdp_in, t1);
1025 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1026 }
1027
1028 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1029 Register reg,
1030 int offset_of_disp) {
1031 assert(ProfileInterpreter, "must be profiling interpreter");
1032 add(t1, mdp_in, reg);
1033 ld(t1, Address(t1, offset_of_disp));
1034 add(mdp_in, mdp_in, t1);
1035 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1036 }
1037
1038
1039 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1040 int constant) {
1041 assert(ProfileInterpreter, "must be profiling interpreter");
1042 addi(mdp_in, mdp_in, (unsigned)constant);
1043 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1044 }
1045
1046
1047 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1048 assert(ProfileInterpreter, "must be profiling interpreter");
1049
1050 // save/restore across call_VM
1051 addi(sp, sp, -2 * wordSize);
1052 sd(zr, Address(sp, 0));
1053 sd(return_bci, Address(sp, wordSize));
1054 call_VM(noreg,
1055 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1056 return_bci);
1057 ld(zr, Address(sp, 0));
1058 ld(return_bci, Address(sp, wordSize));
1059 addi(sp, sp, 2 * wordSize);
1060 }
1061
1062 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1063 Register bumped_count) {
1064 if (ProfileInterpreter) {
1065 Label profile_continue;
1066
1067 // If no method data exists, go to profile_continue.
1068 // Otherwise, assign to mdp
1069 test_method_data_pointer(mdp, profile_continue);
1070
1071 // We are taking a branch. Increment the taken count.
1072 Address data(mdp, in_bytes(JumpData::taken_offset()));
1073 ld(bumped_count, data);
1074 assert(DataLayout::counter_increment == 1,
1075 "flow-free idiom only works with 1");
1076 addi(bumped_count, bumped_count, DataLayout::counter_increment);
1077 Label L;
1078 // eg: bumped_count=0x7fff ffff ffff ffff + 1 < 0. so we use <= 0;
1079 blez(bumped_count, L); // skip store if counter overflow,
1080 sd(bumped_count, data);
1081 bind(L);
1082 // The method data pointer needs to be updated to reflect the new target.
1083 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1084 bind(profile_continue);
1085 }
1086 }
1087
1088 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1089 if (ProfileInterpreter) {
1090 Label profile_continue;
1091
1092 // If no method data exists, go to profile_continue.
1093 test_method_data_pointer(mdp, profile_continue);
1094
1095 // We are taking a branch. Increment the not taken count.
1096 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1097
1098 // The method data pointer needs to be updated to correspond to
1099 // the next bytecode
1100 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1101 bind(profile_continue);
1102 }
1103 }
1104
1105 void InterpreterMacroAssembler::profile_call(Register mdp) {
1106 if (ProfileInterpreter) {
1107 Label profile_continue;
1108
1109 // If no method data exists, go to profile_continue.
1110 test_method_data_pointer(mdp, profile_continue);
1111
1112 // We are making a call. Increment the count.
1113 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1114
1115 // The method data pointer needs to be updated to reflect the new target.
1116 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1117 bind(profile_continue);
1118 }
1119 }
1120
1121 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1122 if (ProfileInterpreter) {
1123 Label profile_continue;
1124
1125 // If no method data exists, go to profile_continue.
1126 test_method_data_pointer(mdp, profile_continue);
1127
1128 // We are making a call. Increment the count.
1129 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1130
1131 // The method data pointer needs to be updated to reflect the new target.
1132 update_mdp_by_constant(mdp,
1133 in_bytes(VirtualCallData::
1134 virtual_call_data_size()));
1135 bind(profile_continue);
1136 }
1137 }
1138
1139
1140 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1141 Register mdp,
1142 Register reg2,
1143 bool receiver_can_be_null) {
1144 if (ProfileInterpreter) {
1145 Label profile_continue;
1146
1147 // If no method data exists, go to profile_continue.
1148 test_method_data_pointer(mdp, profile_continue);
1149
1150 Label skip_receiver_profile;
1151 if (receiver_can_be_null) {
1152 Label not_null;
1153 // We are making a call. Increment the count for null receiver.
1154 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1155 j(skip_receiver_profile);
1156 bind(not_null);
1157 }
1158
1159 // Record the receiver type.
1160 record_klass_in_profile(receiver, mdp, reg2);
1161 bind(skip_receiver_profile);
1162
1163 // The method data pointer needs to be updated to reflect the new target.
1164
1165 update_mdp_by_constant(mdp,
1166 in_bytes(VirtualCallData::
1167 virtual_call_data_size()));
1168 bind(profile_continue);
1169 }
1170 }
1171
1172 // This routine creates a state machine for updating the multi-row
1173 // type profile at a virtual call site (or other type-sensitive bytecode).
1174 // The machine visits each row (of receiver/count) until the receiver type
1175 // is found, or until it runs out of rows. At the same time, it remembers
1176 // the location of the first empty row. (An empty row records null for its
1177 // receiver, and can be allocated for a newly-observed receiver type.)
1178 // Because there are two degrees of freedom in the state, a simple linear
1179 // search will not work; it must be a decision tree. Hence this helper
1180 // function is recursive, to generate the required tree structured code.
1181 // It's the interpreter, so we are trading off code space for speed.
1182 // See below for example code.
1183 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1184 Register receiver, Register mdp,
1185 Register reg2, Label& done) {
1186 if (TypeProfileWidth == 0) {
1187 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1188 } else {
1189 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1190 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1191 }
1192 }
1193
1194 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1195 Register reg2, int start_row, Label& done, int total_rows,
1196 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1197 int last_row = total_rows - 1;
1198 assert(start_row <= last_row, "must be work left to do");
1199 // Test this row for both the item and for null.
1200 // Take any of three different outcomes:
1201 // 1. found item => increment count and goto done
1202 // 2. found null => keep looking for case 1, maybe allocate this cell
1203 // 3. found something else => keep looking for cases 1 and 2
1204 // Case 3 is handled by a recursive call.
1205 for (int row = start_row; row <= last_row; row++) {
1206 Label next_test;
1207 bool test_for_null_also = (row == start_row);
1208
1209 // See if the item is item[n].
1210 int item_offset = in_bytes(item_offset_fn(row));
1211 test_mdp_data_at(mdp, item_offset, item,
1212 (test_for_null_also ? reg2 : noreg),
1213 next_test);
1214 // (Reg2 now contains the item from the CallData.)
1215
1216 // The item is item[n]. Increment count[n].
1217 int count_offset = in_bytes(item_count_offset_fn(row));
1218 increment_mdp_data_at(mdp, count_offset);
1219 j(done);
1220 bind(next_test);
1221
1222 if (test_for_null_also) {
1223 Label found_null;
1224 // Failed the equality check on item[n]... Test for null.
1225 if (start_row == last_row) {
1226 // The only thing left to do is handle the null case.
1227 beqz(reg2, found_null);
1228 // Item did not match any saved item and there is no empty row for it.
1229 // Increment total counter to indicate polymorphic case.
1230 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1231 j(done);
1232 bind(found_null);
1233 break;
1234 }
1235 // Since null is rare, make it be the branch-taken case.
1236 beqz(reg2, found_null);
1237
1238 // Put all the "Case 3" tests here.
1239 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1240 item_offset_fn, item_count_offset_fn);
1241
1242 // Found a null. Keep searching for a matching item,
1243 // but remember that this is an empty (unused) slot.
1244 bind(found_null);
1245 }
1246 }
1247
1248 // In the fall-through case, we found no matching item, but we
1249 // observed the item[start_row] is null.
1250 // Fill in the item field and increment the count.
1251 int item_offset = in_bytes(item_offset_fn(start_row));
1252 set_mdp_data_at(mdp, item_offset, item);
1253 int count_offset = in_bytes(item_count_offset_fn(start_row));
1254 mv(reg2, DataLayout::counter_increment);
1255 set_mdp_data_at(mdp, count_offset, reg2);
1256 if (start_row > 0) {
1257 j(done);
1258 }
1259 }
1260
1261 // Example state machine code for three profile rows:
1262 // # main copy of decision tree, rooted at row[1]
1263 // if (row[0].rec == rec) then [
1264 // row[0].incr()
1265 // goto done
1266 // ]
1267 // if (row[0].rec != nullptr) then [
1268 // # inner copy of decision tree, rooted at row[1]
1269 // if (row[1].rec == rec) then [
1270 // row[1].incr()
1271 // goto done
1272 // ]
1273 // if (row[1].rec != nullptr) then [
1274 // # degenerate decision tree, rooted at row[2]
1275 // if (row[2].rec == rec) then [
1276 // row[2].incr()
1277 // goto done
1278 // ]
1279 // if (row[2].rec != nullptr) then [
1280 // count.incr()
1281 // goto done
1282 // ] # overflow
1283 // row[2].init(rec)
1284 // goto done
1285 // ] else [
1286 // # remember row[1] is empty
1287 // if (row[2].rec == rec) then [
1288 // row[2].incr()
1289 // goto done
1290 // ]
1291 // row[1].init(rec)
1292 // goto done
1293 // ]
1294 // else [
1295 // # remember row[0] is empty
1296 // if (row[1].rec == rec) then [
1297 // row[1].incr()
1298 // goto done
1299 // ]
1300 // if (row[2].rec == rec) then [
1301 // row[2].incr()
1302 // goto done
1303 // ]
1304 // row[0].init(rec)
1305 // goto done
1306 // ]
1307 // done:
1308
1309 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1310 Register mdp, Register reg2) {
1311 assert(ProfileInterpreter, "must be profiling");
1312 Label done;
1313
1314 record_klass_in_profile_helper(receiver, mdp, reg2, done);
1315
1316 bind(done);
1317 }
1318
1319 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1320 if (ProfileInterpreter) {
1321 Label profile_continue;
1322
1323 // If no method data exists, go to profile_continue.
1324 test_method_data_pointer(mdp, profile_continue);
1325
1326 // Update the total ret count.
1327 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1328
1329 for (uint row = 0; row < RetData::row_limit(); row++) {
1330 Label next_test;
1331
1332 // See if return_bci is equal to bci[n]:
1333 test_mdp_data_at(mdp,
1334 in_bytes(RetData::bci_offset(row)),
1335 return_bci, noreg,
1336 next_test);
1337
1338 // return_bci is equal to bci[n]. Increment the count.
1339 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1340
1341 // The method data pointer needs to be updated to reflect the new target.
1342 update_mdp_by_offset(mdp,
1343 in_bytes(RetData::bci_displacement_offset(row)));
1344 j(profile_continue);
1345 bind(next_test);
1346 }
1347
1348 update_mdp_for_ret(return_bci);
1349
1350 bind(profile_continue);
1351 }
1352 }
1353
1354 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1355 if (ProfileInterpreter) {
1356 Label profile_continue;
1357
1358 // If no method data exists, go to profile_continue.
1359 test_method_data_pointer(mdp, profile_continue);
1360
1361 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1362
1363 // The method data pointer needs to be updated.
1364 int mdp_delta = in_bytes(BitData::bit_data_size());
1365 if (TypeProfileCasts) {
1366 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1367 }
1368 update_mdp_by_constant(mdp, mdp_delta);
1369
1370 bind(profile_continue);
1371 }
1372 }
1373
1374 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1375 if (ProfileInterpreter) {
1376 Label profile_continue;
1377
1378 // If no method data exists, go to profile_continue.
1379 test_method_data_pointer(mdp, profile_continue);
1380
1381 // The method data pointer needs to be updated.
1382 int mdp_delta = in_bytes(BitData::bit_data_size());
1383 if (TypeProfileCasts) {
1384 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1385
1386 // Record the object type.
1387 record_klass_in_profile(klass, mdp, reg2);
1388 }
1389 update_mdp_by_constant(mdp, mdp_delta);
1390
1391 bind(profile_continue);
1392 }
1393 }
1394
1395 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1396 if (ProfileInterpreter) {
1397 Label profile_continue;
1398
1399 // If no method data exists, go to profile_continue.
1400 test_method_data_pointer(mdp, profile_continue);
1401
1402 // Update the default case count
1403 increment_mdp_data_at(mdp,
1404 in_bytes(MultiBranchData::default_count_offset()));
1405
1406 // The method data pointer needs to be updated.
1407 update_mdp_by_offset(mdp,
1408 in_bytes(MultiBranchData::
1409 default_displacement_offset()));
1410
1411 bind(profile_continue);
1412 }
1413 }
1414
1415 void InterpreterMacroAssembler::profile_switch_case(Register index,
1416 Register mdp,
1417 Register reg2) {
1418 if (ProfileInterpreter) {
1419 Label profile_continue;
1420
1421 // If no method data exists, go to profile_continue.
1422 test_method_data_pointer(mdp, profile_continue);
1423
1424 // Build the base (index * per_case_size_in_bytes()) +
1425 // case_array_offset_in_bytes()
1426 mv(reg2, in_bytes(MultiBranchData::per_case_size()));
1427 mv(t0, in_bytes(MultiBranchData::case_array_offset()));
1428 Assembler::mul(index, index, reg2);
1429 Assembler::add(index, index, t0);
1430
1431 // Update the case count
1432 increment_mdp_data_at(mdp,
1433 index,
1434 in_bytes(MultiBranchData::relative_count_offset()));
1435
1436 // The method data pointer need to be updated.
1437 update_mdp_by_offset(mdp,
1438 index,
1439 in_bytes(MultiBranchData::
1440 relative_displacement_offset()));
1441
1442 bind(profile_continue);
1443 }
1444 }
1445
1446 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1447
1448 void InterpreterMacroAssembler::notify_method_entry() {
1449 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1450 // track stack depth. If it is possible to enter interp_only_mode we add
1451 // the code to check if the event should be sent.
1452 if (JvmtiExport::can_post_interpreter_events()) {
1453 Label L;
1454 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1455 beqz(x13, L);
1456 call_VM(noreg, CAST_FROM_FN_PTR(address,
1457 InterpreterRuntime::post_method_entry));
1458 bind(L);
1459 }
1460
1461 if (DTraceMethodProbes) {
1462 get_method(c_rarg1);
1463 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1464 xthread, c_rarg1);
1465 }
1466
1467 // RedefineClasses() tracing support for obsolete method entry
1468 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1469 get_method(c_rarg1);
1470 call_VM_leaf(
1471 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1472 xthread, c_rarg1);
1473 }
1474 }
1475
1476
1477 void InterpreterMacroAssembler::notify_method_exit(
1478 TosState state, NotifyMethodExitMode mode) {
1479 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1480 // track stack depth. If it is possible to enter interp_only_mode we add
1481 // the code to check if the event should be sent.
1482 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1483 Label L;
1484 // Note: frame::interpreter_frame_result has a dependency on how the
1485 // method result is saved across the call to post_method_exit. If this
1486 // is changed then the interpreter_frame_result implementation will
1487 // need to be updated too.
1488
1489 // template interpreter will leave the result on the top of the stack.
1490 push(state);
1491 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1492 beqz(x13, L);
1493 call_VM(noreg,
1494 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1495 bind(L);
1496 pop(state);
1497 }
1498
1499 if (DTraceMethodProbes) {
1500 push(state);
1501 get_method(c_rarg1);
1502 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1503 xthread, c_rarg1);
1504 pop(state);
1505 }
1506 }
1507
1508
1509 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1510 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1511 int increment, Address mask,
1512 Register tmp1, Register tmp2,
1513 bool preloaded, Label* where) {
1514 Label done;
1515 if (!preloaded) {
1516 lwu(tmp1, counter_addr);
1517 }
1518 add(tmp1, tmp1, increment);
1519 sw(tmp1, counter_addr);
1520 lwu(tmp2, mask);
1521 andr(tmp1, tmp1, tmp2);
1522 bnez(tmp1, done);
1523 j(*where); // offset is too large so we have to use j instead of beqz here
1524 bind(done);
1525 }
1526
1527 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1528 int number_of_arguments) {
1529 // interpreter specific
1530 //
1531 // Note: No need to save/restore rbcp & rlocals pointer since these
1532 // are callee saved registers and no blocking/ GC can happen
1533 // in leaf calls.
1534 #ifdef ASSERT
1535 {
1536 Label L;
1537 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1538 beqz(t0, L);
1539 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1540 " last_sp isn't null");
1541 bind(L);
1542 }
1543 #endif /* ASSERT */
1544 // super call
1545 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1546 }
1547
1548 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1549 Register java_thread,
1550 Register last_java_sp,
1551 address entry_point,
1552 int number_of_arguments,
1553 bool check_exceptions) {
1554 // interpreter specific
1555 //
1556 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1557 // really make a difference for these runtime calls, since they are
1558 // slow anyway. Btw., bcp must be saved/restored since it may change
1559 // due to GC.
1560 save_bcp();
1561 #ifdef ASSERT
1562 {
1563 Label L;
1564 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1565 beqz(t0, L);
1566 stop("InterpreterMacroAssembler::call_VM_base:"
1567 " last_sp isn't null");
1568 bind(L);
1569 }
1570 #endif /* ASSERT */
1571 // super call
1572 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1573 entry_point, number_of_arguments,
1574 check_exceptions);
1575 // interpreter specific
1576 restore_bcp();
1577 restore_locals();
1578 }
1579
1580 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1581 address entry_point,
1582 Register arg_1) {
1583 assert(arg_1 == c_rarg1, "");
1584 Label resume_pc, not_preempted;
1585
1586 push_cont_fastpath();
1587
1588 // Make VM call. In case of preemption set last_pc to
1589 // the one we want to resume to.
1590 la(t0, resume_pc);
1591 sd(t0, Address(xthread, JavaThread::last_Java_pc_offset()));
1592 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1593
1594 pop_cont_fastpath();
1595
1596 // Check if preempted
1597 ld(t0, Address(xthread, JavaThread::preempt_alternate_return_offset()));
1598 beqz(t0, not_preempted);
1599 sd(zr, Address(xthread, JavaThread::preempt_alternate_return_offset()));
1600 jr(t0);
1601
1602 bind(resume_pc);
1603 restore_after_resume(false /* is_native */);
1604
1605 bind(not_preempted);
1606 }
1607
1608 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1609 la(t0, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1610 jalr(t0);
1611 if (is_native) {
1612 // On resume we need to set up stack as expected
1613 push(dtos);
1614 push(ltos);
1615 }
1616 }
1617
1618 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1619 assert_different_registers(obj, tmp, t0, mdo_addr.base());
1620 Label update, next, none;
1621
1622 verify_oop(obj);
1623
1624 bnez(obj, update);
1625 orptr(mdo_addr, TypeEntries::null_seen, t0, tmp);
1626 j(next);
1627
1628 bind(update);
1629 load_klass(obj, obj);
1630
1631 ld(tmp, mdo_addr);
1632 xorr(obj, obj, tmp);
1633 andi(t0, obj, TypeEntries::type_klass_mask);
1634 beqz(t0, next); // klass seen before, nothing to
1635 // do. The unknown bit may have been
1636 // set already but no need to check.
1637
1638 test_bit(t0, obj, exact_log2(TypeEntries::type_unknown));
1639 bnez(t0, next);
1640 // already unknown. Nothing to do anymore.
1641
1642 beqz(tmp, none);
1643 mv(t0, (u1)TypeEntries::null_seen);
1644 beq(tmp, t0, none);
1645 // There is a chance that the checks above
1646 // fail if another thread has just set the
1647 // profiling to this obj's klass
1648 xorr(obj, obj, tmp); // get back original value before XOR
1649 ld(tmp, mdo_addr);
1650 xorr(obj, obj, tmp);
1651 andi(t0, obj, TypeEntries::type_klass_mask);
1652 beqz(t0, next);
1653
1654 // different than before. Cannot keep accurate profile.
1655 orptr(mdo_addr, TypeEntries::type_unknown, t0, tmp);
1656 j(next);
1657
1658 bind(none);
1659 // first time here. Set profile type.
1660 sd(obj, mdo_addr);
1661 #ifdef ASSERT
1662 andi(obj, obj, TypeEntries::type_mask);
1663 verify_klass_ptr(obj);
1664 #endif
1665
1666 bind(next);
1667 }
1668
1669 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1670 if (!ProfileInterpreter) {
1671 return;
1672 }
1673
1674 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1675 Label profile_continue;
1676
1677 test_method_data_pointer(mdp, profile_continue);
1678
1679 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1680
1681 lbu(t0, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1682 if (is_virtual) {
1683 mv(tmp, (u1)DataLayout::virtual_call_type_data_tag);
1684 bne(t0, tmp, profile_continue);
1685 } else {
1686 mv(tmp, (u1)DataLayout::call_type_data_tag);
1687 bne(t0, tmp, profile_continue);
1688 }
1689
1690 // calculate slot step
1691 static int stack_slot_offset0 = in_bytes(TypeEntriesAtCall::stack_slot_offset(0));
1692 static int slot_step = in_bytes(TypeEntriesAtCall::stack_slot_offset(1)) - stack_slot_offset0;
1693
1694 // calculate type step
1695 static int argument_type_offset0 = in_bytes(TypeEntriesAtCall::argument_type_offset(0));
1696 static int type_step = in_bytes(TypeEntriesAtCall::argument_type_offset(1)) - argument_type_offset0;
1697
1698 if (MethodData::profile_arguments()) {
1699 Label done, loop, loopEnd, profileArgument, profileReturnType;
1700 RegSet pushed_registers;
1701 pushed_registers += x15;
1702 pushed_registers += x16;
1703 pushed_registers += x17;
1704 Register mdo_addr = x15;
1705 Register index = x16;
1706 Register off_to_args = x17;
1707 push_reg(pushed_registers, sp);
1708
1709 mv(off_to_args, in_bytes(TypeEntriesAtCall::args_data_offset()));
1710 mv(t0, TypeProfileArgsLimit);
1711 beqz(t0, loopEnd);
1712
1713 mv(index, zr); // index < TypeProfileArgsLimit
1714 bind(loop);
1715 bgtz(index, profileReturnType);
1716 mv(t0, (int)MethodData::profile_return());
1717 beqz(t0, profileArgument); // (index > 0 || MethodData::profile_return()) == false
1718 bind(profileReturnType);
1719 // If return value type is profiled we may have no argument to profile
1720 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1721 mv(t1, - TypeStackSlotEntries::per_arg_count());
1722 mul(t1, index, t1);
1723 add(tmp, tmp, t1);
1724 mv(t1, TypeStackSlotEntries::per_arg_count());
1725 add(t0, mdp, off_to_args);
1726 blt(tmp, t1, done);
1727
1728 bind(profileArgument);
1729
1730 ld(tmp, Address(callee, Method::const_offset()));
1731 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1732 // stack offset o (zero based) from the start of the argument
1733 // list, for n arguments translates into offset n - o - 1 from
1734 // the end of the argument list
1735 mv(t0, stack_slot_offset0);
1736 mv(t1, slot_step);
1737 mul(t1, index, t1);
1738 add(t0, t0, t1);
1739 add(t0, mdp, t0);
1740 ld(t0, Address(t0));
1741 sub(tmp, tmp, t0);
1742 addi(tmp, tmp, -1);
1743 Address arg_addr = argument_address(tmp);
1744 ld(tmp, arg_addr);
1745
1746 mv(t0, argument_type_offset0);
1747 mv(t1, type_step);
1748 mul(t1, index, t1);
1749 add(t0, t0, t1);
1750 add(mdo_addr, mdp, t0);
1751 Address mdo_arg_addr(mdo_addr, 0);
1752 profile_obj_type(tmp, mdo_arg_addr, t1);
1753
1754 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1755 addi(off_to_args, off_to_args, to_add);
1756
1757 // increment index by 1
1758 addi(index, index, 1);
1759 mv(t1, TypeProfileArgsLimit);
1760 blt(index, t1, loop);
1761 bind(loopEnd);
1762
1763 if (MethodData::profile_return()) {
1764 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1765 addi(tmp, tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1766 }
1767
1768 add(t0, mdp, off_to_args);
1769 bind(done);
1770 mv(mdp, t0);
1771
1772 // unspill the clobbered registers
1773 pop_reg(pushed_registers, sp);
1774
1775 if (MethodData::profile_return()) {
1776 // We're right after the type profile for the last
1777 // argument. tmp is the number of cells left in the
1778 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1779 // if there's a return to profile.
1780 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1781 shadd(mdp, tmp, mdp, tmp, exact_log2(DataLayout::cell_size));
1782 }
1783 sd(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1784 } else {
1785 assert(MethodData::profile_return(), "either profile call args or call ret");
1786 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1787 }
1788
1789 // mdp points right after the end of the
1790 // CallTypeData/VirtualCallTypeData, right after the cells for the
1791 // return value type if there's one
1792
1793 bind(profile_continue);
1794 }
1795 }
1796
1797 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1798 assert_different_registers(mdp, ret, tmp, xbcp, t0, t1);
1799 if (ProfileInterpreter && MethodData::profile_return()) {
1800 Label profile_continue, done;
1801
1802 test_method_data_pointer(mdp, profile_continue);
1803
1804 if (MethodData::profile_return_jsr292_only()) {
1805 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1806
1807 // If we don't profile all invoke bytecodes we must make sure
1808 // it's a bytecode we indeed profile. We can't go back to the
1809 // beginning of the ProfileData we intend to update to check its
1810 // type because we're right after it and we don't known its
1811 // length
1812 Label do_profile;
1813 lbu(t0, Address(xbcp, 0));
1814 mv(tmp, (u1)Bytecodes::_invokedynamic);
1815 beq(t0, tmp, do_profile);
1816 mv(tmp, (u1)Bytecodes::_invokehandle);
1817 beq(t0, tmp, do_profile);
1818 get_method(tmp);
1819 lhu(t0, Address(tmp, Method::intrinsic_id_offset()));
1820 mv(t1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1821 bne(t0, t1, profile_continue);
1822 bind(do_profile);
1823 }
1824
1825 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1826 mv(tmp, ret);
1827 profile_obj_type(tmp, mdo_ret_addr, t1);
1828
1829 bind(profile_continue);
1830 }
1831 }
1832
1833 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2, Register tmp3) {
1834 assert_different_registers(t0, t1, mdp, tmp1, tmp2, tmp3);
1835 if (ProfileInterpreter && MethodData::profile_parameters()) {
1836 Label profile_continue, done;
1837
1838 test_method_data_pointer(mdp, profile_continue);
1839
1840 // Load the offset of the area within the MDO used for
1841 // parameters. If it's negative we're not profiling any parameters
1842 lwu(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1843 srli(tmp2, tmp1, 31);
1844 bnez(tmp2, profile_continue); // i.e. sign bit set
1845
1846 // Compute a pointer to the area for parameters from the offset
1847 // and move the pointer to the slot for the last
1848 // parameters. Collect profiling from last parameter down.
1849 // mdo start + parameters offset + array length - 1
1850 add(mdp, mdp, tmp1);
1851 ld(tmp1, Address(mdp, ArrayData::array_len_offset()));
1852 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1853
1854 Label loop;
1855 bind(loop);
1856
1857 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1858 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1859 int per_arg_scale = exact_log2(DataLayout::cell_size);
1860 add(t0, mdp, off_base);
1861 add(t1, mdp, type_base);
1862
1863 shadd(tmp2, tmp1, t0, tmp2, per_arg_scale);
1864 // load offset on the stack from the slot for this parameter
1865 ld(tmp2, Address(tmp2, 0));
1866 neg(tmp2, tmp2);
1867
1868 // read the parameter from the local area
1869 shadd(tmp2, tmp2, xlocals, tmp2, Interpreter::logStackElementSize);
1870 ld(tmp2, Address(tmp2, 0));
1871
1872 // profile the parameter
1873 shadd(t1, tmp1, t1, t0, per_arg_scale);
1874 Address arg_type(t1, 0);
1875 profile_obj_type(tmp2, arg_type, tmp3);
1876
1877 // go to next parameter
1878 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1879 bgez(tmp1, loop);
1880
1881 bind(profile_continue);
1882 }
1883 }
1884
1885 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1886 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1887 // register "cache" is trashed in next ld, so lets use it as a temporary register
1888 get_cache_index_at_bcp(index, cache, 1, sizeof(u4));
1889 // Get address of invokedynamic array
1890 ld(cache, Address(xcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1891 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1892 slli(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1893 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1894 add(cache, cache, index);
1895 }
1896
1897 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1898 // Get index out of bytecode pointer
1899 get_cache_index_at_bcp(index, cache, bcp_offset, sizeof(u2));
1900 // Take shortcut if the size is a power of 2
1901 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1902 slli(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1903 } else {
1904 mv(cache, sizeof(ResolvedFieldEntry));
1905 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1906 }
1907 // Get address of field entries array
1908 ld(cache, Address(xcpool, ConstantPoolCache::field_entries_offset()));
1909 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1910 add(cache, cache, index);
1911 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1912 membar(MacroAssembler::LoadLoad);
1913 }
1914
1915 void InterpreterMacroAssembler::get_method_counters(Register method,
1916 Register mcs, Label& skip) {
1917 Label has_counters;
1918 ld(mcs, Address(method, Method::method_counters_offset()));
1919 bnez(mcs, has_counters);
1920 call_VM(noreg, CAST_FROM_FN_PTR(address,
1921 InterpreterRuntime::build_method_counters), method);
1922 ld(mcs, Address(method, Method::method_counters_offset()));
1923 beqz(mcs, skip); // No MethodCounters allocated, OutOfMemory
1924 bind(has_counters);
1925 }
1926
1927 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1928 // Get index out of bytecode pointer
1929 get_cache_index_at_bcp(index, cache, bcp_offset, sizeof(u2));
1930 mv(cache, sizeof(ResolvedMethodEntry));
1931 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1932
1933 // Get address of field entries array
1934 ld(cache, Address(xcpool, ConstantPoolCache::method_entries_offset()));
1935 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1936 add(cache, cache, index);
1937 }
1938
1939 #ifdef ASSERT
1940 void InterpreterMacroAssembler::verify_access_flags(Register access_flags, uint32_t flag,
1941 const char* msg, bool stop_by_hit) {
1942 Label L;
1943 test_bit(t0, access_flags, exact_log2(flag));
1944 if (stop_by_hit) {
1945 beqz(t0, L);
1946 } else {
1947 bnez(t0, L);
1948 }
1949 stop(msg);
1950 bind(L);
1951 }
1952
1953 void InterpreterMacroAssembler::verify_frame_setup() {
1954 Label L;
1955 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1956 ld(t0, monitor_block_top);
1957 shadd(t0, t0, fp, t0, LogBytesPerWord);
1958 beq(esp, t0, L);
1959 stop("broken stack frame setup in interpreter");
1960 bind(L);
1961 }
1962 #endif